The fundamental goal of this work is to improve our ability to optimize the overall plan for each patient's treatment course ("comprehensive" treatment course optimization). Current planning methods make use of only a static single instance of the patient's anatomy (typically a treatment planning CT scan), and create an optimized treatment plan based on the clinical therapy prescription, using interactive or inverse planning optimization techniques (for Intensity Modulated Radiation Therapy (IMRT)). However, our knowledge of the patient and treatment goals is not static, but dynamic. Throughout the patient's course of treatment, we obtain new information on geometrical localization, inter- and intra-fraction motion, clinical response, and the precision to which we can predict uncertainties in the data used for planning. To better account for and utiltize this knowledge, the proposed research will develop and evaluate new paradigms for comprehensive optimization of the entire treatment course, evaluating potential improvements over the single-instance planning/optimization techniques which are widely used throughout the radiotherapy community. The project will develop a planning/optimization framework for individualized optimization which incorporates geometric, dosimetric, clinical and biological information as it becomes available (Aim 1), study improvements in single-stage optimization procedures for the multi-criteria problems encountered in therapy planning (Aim 2), and explicitly investigate multi-stage optimization methods for planning and delivery of the entire treatment course (Aim 3). This project will show that development of plan optimization strategies which 1) explicitly account for setup uncertainty and/or organ motion, 2) make use of updated clinical, biological, dosimetric and geometric information to refine the plan, and 3) model and optimize multi-stage adaptive therapy for the entire treatment course will better tailor the overall treatment plan to each individual patient and predict improvements over the current static inverse plan generated once for each patient.